internal static v128 max_uint(v128 a, v128 b)
{
if (Sse4_1.IsSse41Supported)
{
return(Sse4_1.max_epu32(a, b));
}
else if (Sse2.IsSse2Supported)
{
return(Mask.BlendV(a, b, greater_mask_uint(b, a)));
}
else
{
throw new CPUFeatureCheckException();
}
}
public static uint4 gcd(uint4 x, uint4 y)
{
if (Sse2.IsSse2Supported)
{
v128 ZERO = default(v128);
v128 _x = *(v128 *)&x;
v128 _y = *(v128 *)&y;
v128 result = ZERO;
v128 result_if_zero_any = ZERO;
v128 x_is_zero = Sse2.cmpeq_epi32(_x, ZERO);
v128 y_is_zero = Sse2.cmpeq_epi32(_y, ZERO);
v128 any_zero = Sse2.or_si128(x_is_zero, y_is_zero);
result_if_zero_any = Mask.BlendV(result_if_zero_any, _y, x_is_zero);
result_if_zero_any = Mask.BlendV(result_if_zero_any, _x, y_is_zero);
v128 doneMask = any_zero;
int4 shift = math.tzcnt(x | y);
x = shrl(x, math.tzcnt(x));
_x = *(v128 *)&x;
do
{
uint4 temp_y = shrl(*(uint4 *)&_y, math.tzcnt(*(uint4 *)&_y));
_y = *(v128 *)&temp_y;
if (Sse4_1.IsSse41Supported)
{
v128 tempX = _x;
_x = Sse4_1.min_epu32(_x, _y);
_y = Sse4_1.max_epu32(_y, tempX);
}
else
{
v128 tempX = _x;
v128 x_greater_y = Operator.greater_mask_uint(_x, _y);
_x = Mask.BlendV(_x, _y, x_greater_y);
_y = Mask.BlendV(_y, tempX, x_greater_y);
}
_y = Sse2.sub_epi32(_y, _x);
v128 loopCheck = Sse2.andnot_si128(doneMask, Sse2.cmpeq_epi32(_y, ZERO));
result = Mask.BlendV(result, _x, loopCheck);
doneMask = Sse2.or_si128(doneMask, loopCheck);
} while (bitmask32(4 * sizeof(uint)) != Sse2.movemask_epi8(doneMask));
uint4 result_temp = shl(*(uint4 *)&result, shift);
result = *(v128 *)&result_temp;
result = Mask.BlendV(result, result_if_zero_any, any_zero);
return(*(uint4 *)&result);
}
else
{
return(new uint4(gcd(x.x, y.x), gcd(x.y, y.y), gcd(x.z, y.z), gcd(x.w, y.w)));
}
}